Skip to content

runtime-batch.md

Latest commit

 

History

History
61 lines (44 loc) · 1.43 KB

runtime-batch.md

File metadata and controls

61 lines (44 loc) · 1.43 KB

Mneme Runtime Batch Flow

Mneme can now prepare and apply multi-category runtime runs.

This sits above the single-category runtime bridge.

Script

  • scripts/mneme_runtime_batch.py

Purpose

Batch together the four core categories:

  • projects
  • systems
  • decisions
  • incidents

so the runtime can iterate category tasks and then merge one reviewed pack.

Commands

1. Prepare batch tasks

./scripts/mneme_runtime_batch.py prepare-batch \
  --root /path/to/workspace \
  --out /path/to/output/batch

This writes:

  • manifest.json
  • task-projects.json
  • task-systems.json
  • task-decisions.json
  • task-incidents.json

Each task file contains a runtime-ready taskPrompt.

2. Apply returned candidates

./scripts/mneme_runtime_batch.py apply-batch \
  --manifest /path/to/output/batch/manifest.json \
  --candidate-map \
    projects=/path/to/candidate-projects.json \
    systems=/path/to/candidate-systems.json \
    decisions=/path/to/candidate-decisions.json \
    incidents=/path/to/candidate-incidents.json \
  --reviewed-out /path/to/output/reviewed-pack

This does:

  • validate each category result
  • materialize each category output
  • merge all successful categories into one reviewed pack

Why this matters

Single-category automatic dispatch was already proven. This batch layer is the next step toward a full Mneme compile job that builds a serious reviewed pack in one runtime-guided flow.